Embedded DSP · Signal processing · Edge AI
25 years of turning algorithms and research into production-ready embedded software — from reference implementation and fixed-point conversion to DSP, CPU and NPU optimization, integration and long-term support.
You work directly with the senior engineers who do the work.
Audio and voice codecs, enhancement and protection algorithms, and communications signal processing — much of it on ultra-low-power DSPs.
Video codecs and computer vision for low-power, parallel and multi-core processors — in mobile phones, set-top boxes, cameras and vehicles.
Making demanding algorithms fit real silicon: fixed-point implementation, porting and optimization for DSPs, CPUs and hardware accelerators.
Deploying CNN and RNN models on embedded targets and combining them with classical DSP — model integration, fixed-point and quantized inference, DSP/NPU mapping, and latency, memory and power optimization.
The same core team, working together for 25 years
Founded by former Philips and NXP Semiconductors engineers
Our code ships in semiconductor SDKs, phones, cameras and consumer devices
Europe-based, working with R&D teams on three continents
Global semiconductor vendors and the companies building products on their chips: organizations with audited processes, strict coding and review rules, and R&D teams spread across continents. We have worked inside their processes for 25 years.
We don't publish customer names; we're glad to discuss relevant experience under NDA. Associate member of DSP Valley and member of ANIS Romania.
How we work
Our core strength is the path between a research result and a shipping product. We understand both the algorithm and the platform constraints — and we iterate until they meet.
Review the algorithm, inputs, outputs and requirements. We often catch corner cases the original design missed.
MATLAB · paper · standardBuild a clean, modular C reference on an agreed architecture and API, designed from day one to be ported.
C reference + test vectorsAnalyze ranges, choose formats, tailor every operation — and validate against the original algorithm.
fixed-point C referenceProfile, simplify and tune quality against cycles and memory, then port and optimize for the target.
DSP · CPU · NPU codeSupport integration and testing, then maintain and extend the code for the product's lifetime.
SDK release · L3 supportCase studies
Three examples of algorithms we took all the way into commercial products.
A customer's proprietary MATLAB algorithm turned into a fixed-point C library with a clean API, optimized for several processors and supported through integration.
In production in multiple commercial phones and other products — a multi-year collaboration, still ongoing.
Read the case studyProprietary rate-control, motion-estimation and mode-decision schemes designed for parallel encoding across many cores, plus a decoder for large video sizes.
Shipped in the processor vendor's SDK and in commercial surveillance and imaging products.
Read the case studyAudio codecs, radio and sound-processing applications, speech enhancement and BDTI benchmarks, all optimized for a DSP where every cycle costs battery life.
Included in the platform SDK and used by many of the chip's customers, whom we also support.
Read the case studyPlatforms & technologies
DSPs, CPUs, vision processors and accelerators — including a hand in the TriMedia instruction set itself.
Looking for something specific? Our technologies page is searchable — try “fixed-point”, “SIMD” or “LC3”.
All platforms & technologiesHow we engage
We can rapidly allocate senior engineering capacity — from one specialist inside your team to a complete, defined work package.
A senior DSP or embedded engineer joins your team, working in your repositories, tools and process.
best for: a skills gap on a critical pathA small, self-directed team takes ownership of a component or subsystem, end to end.
best for: a codec, library or pipelineAgreed scope, deliverables and acceptance criteria: a port, an optimization target, a validated component.
best for: clear specs, fixed outcomesA short, focused study: will it run in real time on this target, and what will it take? Profiling, benchmarks, a plan.
best for: go / no-go decisionsReactive maintenance, level-3 support and training for code in the field — including support for your own customers.
best for: products with a long lifeTell us about the target, the budget and the deadline. You will be talking to a senior engineer from the first call.